The methodology for the average should be, and probably is, to determine the number of those polled who favor each candidate by taking the percentage of the number sampled, and summing that over all the polls. Those results are then divided by the total number sampled. Nothing wrong with that approach.
But there are two types of populations reported as sampled. Two-thirds of those sampled are simply registered voters. Rasmussen and two other organizations restrict their samples to likely voters. In New Mexico, the turnout of registered voters in presidential years is about 70%. A registered voter poll is fine if you believe that the preference you are asking about is independent of the likeliness to vote.
But it's not independent in the RCP data. It is clear from a moment's inspection that the likely-voter polls show a much weaker preference for Obama than the registered-voter polls. Furthermore, it's an observation of political polling that the voting electorate is more conservative than the registered electorate.
By separating the three likely-voter polls from the four registered-voter polls for which a sample size is reported, I conclude that while the combined registered-voter sample prefers Obama by 2.6 percentage points while the combined likely-voter sample prefers Romney by 1.6 percentage points, for a net difference of 4.1 percentage points.
Obama
Romney
Undecided
MargOfErr
Since there are about 5600 in the registered-voter sample and 3300 in the likely-voter sample the margin of error is somewhat smaller in the registered-voter sample.
When the Gallup poll switched from sampling registered voters to sampling likely voters in the last months of the 2010 campaign, the support for Republicans in the generic congressional question jumped by over 6 percentage points.
It is no wonder that the Obama campaign is flogging its base, and the tingly-legged punditry is screeching in fear.
